Gait Prediction Of Lower Limb Exoskeleton Robot For Hemiplegia Rehabilitatio

As an advanced rehabilitation medical device,the lower extremity rehabilitation exoskeleton can assist hemiplegic patients to conduct partial rehabilitation training under the guidance of rehabilitation physicians.However,the current lower extremity rehabilitation exoskeleton still has some problems,such as it can not generate appropriate training gait for patients’ training needs in different rehabilitation training modes.In order to overcome the shortcomings of gait prediction in hemiplegic lower extremity exoskeleton,three gait prediction algorithms are proposed in this paper.Firstly,a gait prediction algorithm based on random forest and probabilistic motion primitives is proposed to address the problems of single training gait and poor tunability in the existing passive training for hemiplegia rehabilitation.The probabilistic motion primitives are used to establish the base model of gait trajectory,and the random forest algorithm are used to predict the gait feature points that match the physical characteristics of patients for training gait adjustment.The experimental results show that the average absolute errors of hip and knee joints predicted by the algorithm in this paper are 4.26° and 5.02°,respectively,which effectively reduce the gait prediction errors while ensuring that the training gait can be adjusted.Secondly,a gait phase prediction algorithm based on lower limb joint angle data is designed to address the problem of difficult gait phase prediction due to abnormal plantar pressure data of hemiplegic patients in active training for hemiplegic rehabilitation.The algorithm uses random forest classification to reduce the impact of unbalanced samples on phase prediction accuracy,while input features are filtered by combining random forest and pearson correlation coefficients in order to reduce model complexity.The experimental results show that the average accuracy of gait phase prediction is 91.6%,which meets the need of gait phase prediction in the case of plantar pressure data failure.Finally,a collaborative gait prediction algorithm is designed to address the problem of underutilization of information from the healthy side of the patient in the active training of hemiplegic rehabilitation.The algorithm combines the spatial feature extraction ability of convolutional neural network and the temporal feature extraction ability of long short-term memory network to predict the gait angle of the hemiplegic side with the historical gait data of the healthy side to improve the active participation of the patient in the training process.The experimental results show that the algorithm in this paper achieves lower prediction error,which can achieve better continuous gait prediction on the hemiplegic side.